Nested Named Entity Recognition as Building Local Hypergraphs

Authors: Yukun Yan, Bingling Cai, Sen Song

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments illustrate that our model outperforms previous state-of-the-art methods on four widely used nested named entity recognition datasets: ACE04, ACE05, GENIA, and KBP17.
Researcher Affiliation Academia Yukun Yan,1,2 Bingling Cai,1,2 Sen Song1,2* 1Biomedical Department, Tsinghua University 2Laboratory of Brain and Intelligence, Tsinghua University {yanyk13, caibl13}@mails.tsinghua.edu.cn, songsen@tsinghua.edu.cn
Pseudocode No The paper describes steps and rules within the text and using equations (e.g., rules R for hypergraph building), but it does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block.
Open Source Code Yes The code is available at https://github.com/yanyk13/local-hypergraphbuilding-network.git.
Open Datasets Yes To evaluate the proposed method, we conduct experiments on four widely used datasets for Nested NER: ACE04, ACE05, KBP17 and GENIA. ACE04 and ACE05(Doddington et al. 2004; Stephanie Strassel and Maeda 2006) are nested datasets with 7 entity categories, we use the same setup as previous works(Katiyar and Cardie 2018; Shen et al. 2021) and split them into train, dev, and test sets by 8:1:1. GENIA(Ohta et al. 2002) is a nested dataset consisting of biology texts. There are 5 entity types: DNA, RNA, protein, cell line and cell categories. Following (Shen et al. 2021), we use a 90%/10% train/test split. KBP17(Ji et al. 2017) has 5 entity categories. We split all the samples into 866/20/167 documents for train/dev/test set following the same setup as previous works(Shen et al. 2021).
Dataset Splits Yes ACE04 and ACE05...we use the same setup as previous works(Katiyar and Cardie 2018; Shen et al. 2021) and split them into train, dev, and test sets by 8:1:1. GENIA...we use a 90%/10% train/test split. KBP17...We split all the samples into 866/20/167 documents for train/dev/test set following the same setup as previous works(Shen et al. 2021).
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. It only mentions using BERT-large and Bio BERT-large models, which implies computational resources but without specific hardware identification.
Software Dependencies No The paper mentions specific tools and models like BERT-large, GloVe, Bio BERT-large, Bio Wordvec, and AdamW optimizer, but it does not provide specific version numbers for these or other software dependencies like Python, PyTorch, or TensorFlow.
Experiment Setup Yes Based on the performance on the dev sets of ACE04, ACE05, and KBP17, γ used in equation (6) is set to 0.9, the scale hyper-parameter λ for sampling boundary candidates is set to 5, and the merging threshold θ is set to 0.5. For all the experiments, we train our model for 100 epochs with an Adam W optimizer and a linear warmup-decay learning rate. The initial learning rate for BERT modules and other parameters are set to 1e-5, and 1e-3 respectively.